Merge branch 'master' of /usr/src/ntfs-2.6/
[linux-2.6] / arch / powerpc / mm / mem.c
1 /*
2  *  PowerPC version
3  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
4  *
5  *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
6  *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
7  *    Copyright (C) 1996 Paul Mackerras
8  *  Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
9  *  PPC44x/36-bit changes by Matt Porter (mporter@mvista.com)
10  *
11  *  Derived from "arch/i386/mm/init.c"
12  *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
13  *
14  *  This program is free software; you can redistribute it and/or
15  *  modify it under the terms of the GNU General Public License
16  *  as published by the Free Software Foundation; either version
17  *  2 of the License, or (at your option) any later version.
18  *
19  */
20
21 #include <linux/config.h>
22 #include <linux/module.h>
23 #include <linux/sched.h>
24 #include <linux/kernel.h>
25 #include <linux/errno.h>
26 #include <linux/string.h>
27 #include <linux/types.h>
28 #include <linux/mm.h>
29 #include <linux/stddef.h>
30 #include <linux/init.h>
31 #include <linux/bootmem.h>
32 #include <linux/highmem.h>
33 #include <linux/initrd.h>
34 #include <linux/pagemap.h>
35
36 #include <asm/pgalloc.h>
37 #include <asm/prom.h>
38 #include <asm/io.h>
39 #include <asm/mmu_context.h>
40 #include <asm/pgtable.h>
41 #include <asm/mmu.h>
42 #include <asm/smp.h>
43 #include <asm/machdep.h>
44 #include <asm/btext.h>
45 #include <asm/tlb.h>
46 #include <asm/prom.h>
47 #include <asm/lmb.h>
48 #include <asm/sections.h>
49 #ifdef CONFIG_PPC64
50 #include <asm/vdso.h>
51 #endif
52
53 #include "mmu_decl.h"
54
55 #ifndef CPU_FTR_COHERENT_ICACHE
56 #define CPU_FTR_COHERENT_ICACHE 0       /* XXX for now */
57 #define CPU_FTR_NOEXECUTE       0
58 #endif
59
60 int init_bootmem_done;
61 int mem_init_done;
62 unsigned long memory_limit;
63
64 /*
65  * This is called by /dev/mem to know if a given address has to
66  * be mapped non-cacheable or not
67  */
68 int page_is_ram(unsigned long pfn)
69 {
70         unsigned long paddr = (pfn << PAGE_SHIFT);
71
72 #ifndef CONFIG_PPC64    /* XXX for now */
73         return paddr < __pa(high_memory);
74 #else
75         int i;
76         for (i=0; i < lmb.memory.cnt; i++) {
77                 unsigned long base;
78
79                 base = lmb.memory.region[i].base;
80
81                 if ((paddr >= base) &&
82                         (paddr < (base + lmb.memory.region[i].size))) {
83                         return 1;
84                 }
85         }
86
87         return 0;
88 #endif
89 }
90 EXPORT_SYMBOL(page_is_ram);
91
92 pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
93                               unsigned long size, pgprot_t vma_prot)
94 {
95         if (ppc_md.phys_mem_access_prot)
96                 return ppc_md.phys_mem_access_prot(file, pfn, size, vma_prot);
97
98         if (!page_is_ram(pfn))
99                 vma_prot = __pgprot(pgprot_val(vma_prot)
100                                     | _PAGE_GUARDED | _PAGE_NO_CACHE);
101         return vma_prot;
102 }
103 EXPORT_SYMBOL(phys_mem_access_prot);
104
105 #ifdef CONFIG_MEMORY_HOTPLUG
106
107 void online_page(struct page *page)
108 {
109         ClearPageReserved(page);
110         free_cold_page(page);
111         totalram_pages++;
112         num_physpages++;
113 }
114
115 /*
116  * This works only for the non-NUMA case.  Later, we'll need a lookup
117  * to convert from real physical addresses to nid, that doesn't use
118  * pfn_to_nid().
119  */
120 int __devinit add_memory(u64 start, u64 size)
121 {
122         struct pglist_data *pgdata = NODE_DATA(0);
123         struct zone *zone;
124         unsigned long start_pfn = start >> PAGE_SHIFT;
125         unsigned long nr_pages = size >> PAGE_SHIFT;
126
127         /* this should work for most non-highmem platforms */
128         zone = pgdata->node_zones;
129
130         return __add_pages(zone, start_pfn, nr_pages);
131
132         return 0;
133 }
134
135 /*
136  * First pass at this code will check to determine if the remove
137  * request is within the RMO.  Do not allow removal within the RMO.
138  */
139 int __devinit remove_memory(u64 start, u64 size)
140 {
141         struct zone *zone;
142         unsigned long start_pfn, end_pfn, nr_pages;
143
144         start_pfn = start >> PAGE_SHIFT;
145         nr_pages = size >> PAGE_SHIFT;
146         end_pfn = start_pfn + nr_pages;
147
148         printk("%s(): Attempting to remove memoy in range "
149                         "%lx to %lx\n", __func__, start, start+size);
150         /*
151          * check for range within RMO
152          */
153         zone = page_zone(pfn_to_page(start_pfn));
154
155         printk("%s(): memory will be removed from "
156                         "the %s zone\n", __func__, zone->name);
157
158         /*
159          * not handling removing memory ranges that
160          * overlap multiple zones yet
161          */
162         if (end_pfn > (zone->zone_start_pfn + zone->spanned_pages))
163                 goto overlap;
164
165         /* make sure it is NOT in RMO */
166         if ((start < lmb.rmo_size) || ((start+size) < lmb.rmo_size)) {
167                 printk("%s(): range to be removed must NOT be in RMO!\n",
168                         __func__);
169                 goto in_rmo;
170         }
171
172         return __remove_pages(zone, start_pfn, nr_pages);
173
174 overlap:
175         printk("%s(): memory range to be removed overlaps "
176                 "multiple zones!!!\n", __func__);
177 in_rmo:
178         return -1;
179 }
180 #endif /* CONFIG_MEMORY_HOTPLUG */
181
182 void show_mem(void)
183 {
184         unsigned long total = 0, reserved = 0;
185         unsigned long shared = 0, cached = 0;
186         unsigned long highmem = 0;
187         struct page *page;
188         pg_data_t *pgdat;
189         unsigned long i;
190
191         printk("Mem-info:\n");
192         show_free_areas();
193         printk("Free swap:       %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
194         for_each_pgdat(pgdat) {
195                 unsigned long flags;
196                 pgdat_resize_lock(pgdat, &flags);
197                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
198                         page = pgdat_page_nr(pgdat, i);
199                         total++;
200                         if (PageHighMem(page))
201                                 highmem++;
202                         if (PageReserved(page))
203                                 reserved++;
204                         else if (PageSwapCache(page))
205                                 cached++;
206                         else if (page_count(page))
207                                 shared += page_count(page) - 1;
208                 }
209                 pgdat_resize_unlock(pgdat, &flags);
210         }
211         printk("%ld pages of RAM\n", total);
212 #ifdef CONFIG_HIGHMEM
213         printk("%ld pages of HIGHMEM\n", highmem);
214 #endif
215         printk("%ld reserved pages\n", reserved);
216         printk("%ld pages shared\n", shared);
217         printk("%ld pages swap cached\n", cached);
218 }
219
220 /*
221  * Initialize the bootmem system and give it all the memory we
222  * have available.  If we are using highmem, we only put the
223  * lowmem into the bootmem system.
224  */
225 #ifndef CONFIG_NEED_MULTIPLE_NODES
226 void __init do_init_bootmem(void)
227 {
228         unsigned long i;
229         unsigned long start, bootmap_pages;
230         unsigned long total_pages;
231         int boot_mapsize;
232
233         max_pfn = total_pages = lmb_end_of_DRAM() >> PAGE_SHIFT;
234 #ifdef CONFIG_HIGHMEM
235         total_pages = total_lowmem >> PAGE_SHIFT;
236 #endif
237
238         /*
239          * Find an area to use for the bootmem bitmap.  Calculate the size of
240          * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
241          * Add 1 additional page in case the address isn't page-aligned.
242          */
243         bootmap_pages = bootmem_bootmap_pages(total_pages);
244
245         start = lmb_alloc(bootmap_pages << PAGE_SHIFT, PAGE_SIZE);
246         BUG_ON(!start);
247
248         boot_mapsize = init_bootmem(start >> PAGE_SHIFT, total_pages);
249
250         /* Add all physical memory to the bootmem map, mark each area
251          * present.
252          */
253         for (i = 0; i < lmb.memory.cnt; i++) {
254                 unsigned long base = lmb.memory.region[i].base;
255                 unsigned long size = lmb_size_bytes(&lmb.memory, i);
256 #ifdef CONFIG_HIGHMEM
257                 if (base >= total_lowmem)
258                         continue;
259                 if (base + size > total_lowmem)
260                         size = total_lowmem - base;
261 #endif
262                 free_bootmem(base, size);
263         }
264
265         /* reserve the sections we're already using */
266         for (i = 0; i < lmb.reserved.cnt; i++)
267                 reserve_bootmem(lmb.reserved.region[i].base,
268                                 lmb_size_bytes(&lmb.reserved, i));
269
270         /* XXX need to clip this if using highmem? */
271         for (i = 0; i < lmb.memory.cnt; i++)
272                 memory_present(0, lmb_start_pfn(&lmb.memory, i),
273                                lmb_end_pfn(&lmb.memory, i));
274         init_bootmem_done = 1;
275 }
276
277 /*
278  * paging_init() sets up the page tables - in fact we've already done this.
279  */
280 void __init paging_init(void)
281 {
282         unsigned long zones_size[MAX_NR_ZONES];
283         unsigned long zholes_size[MAX_NR_ZONES];
284         unsigned long total_ram = lmb_phys_mem_size();
285         unsigned long top_of_ram = lmb_end_of_DRAM();
286
287 #ifdef CONFIG_HIGHMEM
288         map_page(PKMAP_BASE, 0, 0);     /* XXX gross */
289         pkmap_page_table = pte_offset_kernel(pmd_offset(pgd_offset_k
290                         (PKMAP_BASE), PKMAP_BASE), PKMAP_BASE);
291         map_page(KMAP_FIX_BEGIN, 0, 0); /* XXX gross */
292         kmap_pte = pte_offset_kernel(pmd_offset(pgd_offset_k
293                         (KMAP_FIX_BEGIN), KMAP_FIX_BEGIN), KMAP_FIX_BEGIN);
294         kmap_prot = PAGE_KERNEL;
295 #endif /* CONFIG_HIGHMEM */
296
297         printk(KERN_INFO "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
298                top_of_ram, total_ram);
299         printk(KERN_INFO "Memory hole size: %ldMB\n",
300                (top_of_ram - total_ram) >> 20);
301         /*
302          * All pages are DMA-able so we put them all in the DMA zone.
303          */
304         memset(zones_size, 0, sizeof(zones_size));
305         memset(zholes_size, 0, sizeof(zholes_size));
306
307         zones_size[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
308         zholes_size[ZONE_DMA] = (top_of_ram - total_ram) >> PAGE_SHIFT;
309
310 #ifdef CONFIG_HIGHMEM
311         zones_size[ZONE_DMA] = total_lowmem >> PAGE_SHIFT;
312         zones_size[ZONE_HIGHMEM] = (total_memory - total_lowmem) >> PAGE_SHIFT;
313         zholes_size[ZONE_HIGHMEM] = (top_of_ram - total_ram) >> PAGE_SHIFT;
314 #else
315         zones_size[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
316         zholes_size[ZONE_DMA] = (top_of_ram - total_ram) >> PAGE_SHIFT;
317 #endif /* CONFIG_HIGHMEM */
318
319         free_area_init_node(0, NODE_DATA(0), zones_size,
320                             __pa(PAGE_OFFSET) >> PAGE_SHIFT, zholes_size);
321 }
322 #endif /* ! CONFIG_NEED_MULTIPLE_NODES */
323
324 void __init mem_init(void)
325 {
326 #ifdef CONFIG_NEED_MULTIPLE_NODES
327         int nid;
328 #endif
329         pg_data_t *pgdat;
330         unsigned long i;
331         struct page *page;
332         unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
333
334         num_physpages = max_pfn;        /* RAM is assumed contiguous */
335         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
336
337 #ifdef CONFIG_NEED_MULTIPLE_NODES
338         for_each_online_node(nid) {
339                 if (NODE_DATA(nid)->node_spanned_pages != 0) {
340                         printk("freeing bootmem node %x\n", nid);
341                         totalram_pages +=
342                                 free_all_bootmem_node(NODE_DATA(nid));
343                 }
344         }
345 #else
346         max_mapnr = num_physpages;
347         totalram_pages += free_all_bootmem();
348 #endif
349         for_each_pgdat(pgdat) {
350                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
351                         page = pgdat_page_nr(pgdat, i);
352                         if (PageReserved(page))
353                                 reservedpages++;
354                 }
355         }
356
357         codesize = (unsigned long)&_sdata - (unsigned long)&_stext;
358         datasize = (unsigned long)&__init_begin - (unsigned long)&_sdata;
359         initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
360         bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
361
362 #ifdef CONFIG_HIGHMEM
363         {
364                 unsigned long pfn, highmem_mapnr;
365
366                 highmem_mapnr = total_lowmem >> PAGE_SHIFT;
367                 for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
368                         struct page *page = pfn_to_page(pfn);
369
370                         ClearPageReserved(page);
371                         set_page_count(page, 1);
372                         __free_page(page);
373                         totalhigh_pages++;
374                 }
375                 totalram_pages += totalhigh_pages;
376                 printk(KERN_INFO "High memory: %luk\n",
377                        totalhigh_pages << (PAGE_SHIFT-10));
378         }
379 #endif /* CONFIG_HIGHMEM */
380
381         printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
382                "%luk reserved, %luk data, %luk bss, %luk init)\n",
383                 (unsigned long)nr_free_pages() << (PAGE_SHIFT-10),
384                 num_physpages << (PAGE_SHIFT-10),
385                 codesize >> 10,
386                 reservedpages << (PAGE_SHIFT-10),
387                 datasize >> 10,
388                 bsssize >> 10,
389                 initsize >> 10);
390
391         mem_init_done = 1;
392
393 #ifdef CONFIG_PPC64
394         /* Initialize the vDSO */
395         vdso_init();
396 #endif
397 }
398
399 /*
400  * This is called when a page has been modified by the kernel.
401  * It just marks the page as not i-cache clean.  We do the i-cache
402  * flush later when the page is given to a user process, if necessary.
403  */
404 void flush_dcache_page(struct page *page)
405 {
406         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
407                 return;
408         /* avoid an atomic op if possible */
409         if (test_bit(PG_arch_1, &page->flags))
410                 clear_bit(PG_arch_1, &page->flags);
411 }
412 EXPORT_SYMBOL(flush_dcache_page);
413
414 void flush_dcache_icache_page(struct page *page)
415 {
416 #ifdef CONFIG_BOOKE
417         void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
418         __flush_dcache_icache(start);
419         kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
420 #elif defined(CONFIG_8xx) || defined(CONFIG_PPC64)
421         /* On 8xx there is no need to kmap since highmem is not supported */
422         __flush_dcache_icache(page_address(page)); 
423 #else
424         __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
425 #endif
426
427 }
428 void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
429 {
430         clear_page(page);
431
432         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
433                 return;
434         /*
435          * We shouldnt have to do this, but some versions of glibc
436          * require it (ld.so assumes zero filled pages are icache clean)
437          * - Anton
438          */
439
440         /* avoid an atomic op if possible */
441         if (test_bit(PG_arch_1, &pg->flags))
442                 clear_bit(PG_arch_1, &pg->flags);
443 }
444 EXPORT_SYMBOL(clear_user_page);
445
446 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
447                     struct page *pg)
448 {
449         copy_page(vto, vfrom);
450
451         /*
452          * We should be able to use the following optimisation, however
453          * there are two problems.
454          * Firstly a bug in some versions of binutils meant PLT sections
455          * were not marked executable.
456          * Secondly the first word in the GOT section is blrl, used
457          * to establish the GOT address. Until recently the GOT was
458          * not marked executable.
459          * - Anton
460          */
461 #if 0
462         if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
463                 return;
464 #endif
465
466         if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
467                 return;
468
469         /* avoid an atomic op if possible */
470         if (test_bit(PG_arch_1, &pg->flags))
471                 clear_bit(PG_arch_1, &pg->flags);
472 }
473
474 void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
475                              unsigned long addr, int len)
476 {
477         unsigned long maddr;
478
479         maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
480         flush_icache_range(maddr, maddr + len);
481         kunmap(page);
482 }
483 EXPORT_SYMBOL(flush_icache_user_range);
484
485 /*
486  * This is called at the end of handling a user page fault, when the
487  * fault has been handled by updating a PTE in the linux page tables.
488  * We use it to preload an HPTE into the hash table corresponding to
489  * the updated linux PTE.
490  * 
491  * This must always be called with the mm->page_table_lock held
492  */
493 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
494                       pte_t pte)
495 {
496         /* handle i-cache coherency */
497         unsigned long pfn = pte_pfn(pte);
498 #ifdef CONFIG_PPC32
499         pmd_t *pmd;
500 #else
501         unsigned long vsid;
502         void *pgdir;
503         pte_t *ptep;
504         int local = 0;
505         cpumask_t tmp;
506         unsigned long flags;
507 #endif
508
509         /* handle i-cache coherency */
510         if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE) &&
511             !cpu_has_feature(CPU_FTR_NOEXECUTE) &&
512             pfn_valid(pfn)) {
513                 struct page *page = pfn_to_page(pfn);
514                 if (!PageReserved(page)
515                     && !test_bit(PG_arch_1, &page->flags)) {
516                         if (vma->vm_mm == current->active_mm) {
517 #ifdef CONFIG_8xx
518                         /* On 8xx, cache control instructions (particularly 
519                          * "dcbst" from flush_dcache_icache) fault as write 
520                          * operation if there is an unpopulated TLB entry 
521                          * for the address in question. To workaround that, 
522                          * we invalidate the TLB here, thus avoiding dcbst 
523                          * misbehaviour.
524                          */
525                                 _tlbie(address);
526 #endif
527                                 __flush_dcache_icache((void *) address);
528                         } else
529                                 flush_dcache_icache_page(page);
530                         set_bit(PG_arch_1, &page->flags);
531                 }
532         }
533
534 #ifdef CONFIG_PPC_STD_MMU
535         /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
536         if (!pte_young(pte) || address >= TASK_SIZE)
537                 return;
538 #ifdef CONFIG_PPC32
539         if (Hash == 0)
540                 return;
541         pmd = pmd_offset(pgd_offset(vma->vm_mm, address), address);
542         if (!pmd_none(*pmd))
543                 add_hash_page(vma->vm_mm->context, address, pmd_val(*pmd));
544 #else
545         pgdir = vma->vm_mm->pgd;
546         if (pgdir == NULL)
547                 return;
548
549         ptep = find_linux_pte(pgdir, address);
550         if (!ptep)
551                 return;
552
553         vsid = get_vsid(vma->vm_mm->context.id, address);
554
555         local_irq_save(flags);
556         tmp = cpumask_of_cpu(smp_processor_id());
557         if (cpus_equal(vma->vm_mm->cpu_vm_mask, tmp))
558                 local = 1;
559
560         __hash_page(address, 0, vsid, ptep, 0x300, local);
561         local_irq_restore(flags);
562 #endif
563 #endif
564 }